Paper Fact Checker
Paste the following prompt into your AI chat to install this skill:
Please install @user_d0deb03a/paper-fact-checker according to https://skillhub.cn/install/skillhub.md.
About this skill
Problem
Papers often present unverified numbers, inflated comparisons, or invented references as facts. Manual checking is expensive, and plausible-looking citations can hide missing or mismatched sources. This skill audits verifiable claims without judging novelty, writing, or academic value.
How It Works
- After receiving
Markdown,PDF, or plain text, it checks for empty input, format, length, and prompt-injection phrases before extracting claims. - It extracts up to
30high-value claims by default, covering numeric, comparative, citation, factual, and methodological claims, then asks for confirmation before searching. - Evidence is selected by claim type: official docs, primary papers, benchmark leaderboards, open databases, or code repositories; each entry records title,
URL, and retrieval date. - References are checked separately for existence, citation accuracy, and locateable links; unverifiable references are flagged as possible hallucinations.
- The final
Markdownreport includes an overall score, claim tables, and priority issues. Scores reflect evidence strength only, and verification limits are preserved when the claim is outside the verifiable scope.
Use Cases
- Before submission, audit dataset size, model metrics, and citations in a paper to produce a verifiable report.
- During peer review, check whether cited DOIs exist and whether cited conclusions are supported by the original source.
- Before reproduction, verify hyperparameters, optimizers, and benchmark names against official documentation.
- Before writing a survey, confirm data comparisons, SOTA claims, and reference links are locateable across papers.
Best For
- Graduate students submitting ML papers who need to verify experimental numbers and locateable citations item by item.
- Conference paper editors who need to detect fake DOIs, exaggerated citations, and data inconsistencies before review.
- Researchers writing literature surveys who need to confirm benchmark names, metric direction, and source links across papers.
- Engineers reproducing third-party results who need to verify optimizers, hyperparameters, and model names against official docs.
Related Skills
Search Huawei Cloud official docs and product pages to find ECS, OBS, RDS, CCE product specs, parameters, documentation, and API references without login.
OCR-based recognition for movie, train, flight, and event tickets in images or PDFs, extracting key fields into Markdown or JSON reports.
Turns notes, research, and meeting summaries into actionable next moves, plans, decisions, experiments, and decision-changing gaps.
A local wiki knowledge base manager that compiles raw documents into sourced, indexed Markdown pages with wikilinks, query support, and health checks.